EEG Complexity as a Biomarker for Autism Spectrum Disorder Risk Personnel William Bosl (PI) Instructor, Children's Hospital Informatics Program (ChIP) Charles Nelson (collaborator) Professor, Harvard and Division of Developmental Medicine at CHB Helen Tager-Flusberg (consultant) Professor, Boston University Department of Psychology Abstract Autism spectrum disorders (ASD) are complex, heritable disorders that have highly variable long-term outcomes. Research suggests that complex mental disorders such as autism are associated with abnormal brain connectivity that may vary between different regions and different scales [1]. In the autistic brain, high local connectivity and low long-range connectivity may develop concurrently due to problems with synapse pruning or formation [1, 2]. Estimation of changes in neural connectivity might be an effective diagnostic biomarker for abnormal connectivity development that leads to ASD behaviors. The electrical signals produced by neural networks contain information about the network structure [3-5]. Nonlinear signal processing algorithms [6, 7] can be used to compute features from the time series produced by the network to characterize the dynamics of a complex system such as the brain. Our hypothesis is that multiscale entropy (MSE) is one measure of signal complexity that will reveal distinct, measurable differences between normally developing brains and those that will eventually be diagnosed with an ASD. Machine learning algorithms will be used to classify infants into one of three groups using MSE values: controls (CON), high risk based on having an older sibling with autism, but does not develop autism (HRN) and high risk and develops autism (HRA). Rather than mapping to these three risk groups, MSE values will be mapped to a severity score based on Autism Diagnostic Observation Schedule (ADOS) or equivalent assessments given to all infants in the study. Preliminary data using MSE to measure signal complexity shows a clear difference between normal controls and a group of infants at high risk for developing ASD based on family history (Bosl, et al., 2011). The change was particularly striking between 9 and 12 months of age when critical cognitive milestones are expected in normal infants. In the proposed study, we will also attempt to make predictions of outcome using MSE growth trajectories, with feature vectors composed of all values up to a specified age: 6-9, 6-12, 6-18 and 6-24 months. We hypothesize that the growth trajectories may be more informative than measures at one given age. To establish a baseline for comparison of predictions based on MSE, a similar prediction calculation using machine learning algorithms will be done with all available assessment scores, including ADOS, SCQ, Mullen and physiological measures such as head circumference. Even if a diagnosis of autism is not possible from assessment scores at one age before 18 months, it may be that the trajectory of assessment scores has predictive value. In all prediction computations, an estimate will be made of which variables contribute the most information to the prediction (whether EEG channels or assessment scores). Finally, correlations between MSE values and assessment scores will be determined in order to judge whether MSE measurements are proxies for assessment scores, or contribute complementary information, or neither. Early diagnosis and therapy are known to significantly improve the long-term prognosis of ASD patients. This project has the potential to enable early diagnosis of ASD, within the first year of life, and assess severity. If successful, this will also enable a new class of therapies aimed at averting the development of autistic brain functional tendencies before they are fully formed. The novel methodology developed here, using complex systems methods to extract signal features and machine learning to map MSE values, traditional scores and physical measurements to autism severity estimates, may be widely applicable as an approach for identifying quantitative biomarkers of other mental disorders. This may have particular value in resource poor regions where few professionals are available for complete behavioral assessments. PHS 398/2590 (Rev. 11/07) Page Continuation Format Page PUBLIC HEALTH RELEVANCE: Autism spectrum disorders are a complex and heterogeneous set of disorders that affect the developmental trajectory in several key behavioral domains including social, cognitive and language. The underlying brain dysfunction that results in the behavioral characteristics is not well understood. Because the proposed grant will leverage data from existing grants, it is uniquely positioned to explore novel diagnostic methods in a longitudinal study in a much shorter time than would normally be required. To our knowledge, our preliminary results are the first demonstration of the use of an information-theoretic value derived from EEG data, the multiscale entropy, as a biomarker for infants at risk for a complex neurodevelopmental disorder. The novel methodology developed here, using complex systems methods to extract signal features and the use of machine learning to discover patterns in signal features may not only enable early diagnosis of autism, but also provide a means of monitoring the effectiveness of early interventions.